Examining antisemitic and anti-Muslim hate speech on Canadian YouTube channels in the aftermath of 7 October

17 June 2024

This article summarises the findings of a topic modelling analysis of antisemitic and anti-Muslim hate speech on Canadian YouTube comments in the aftermath of the 7 October attack. It is part of a research series looking at the impacts of the Israel-Hamas conflict on extremism, hate and disinformation in Canada. This project has been made possible in part by the Government of Canada. All views are ISD’s own.


Key Findings

  • Previous ISD research identified a 50-fold increase in the volume of antisemitic comments and a 43-fold increase in the volume of anti-Muslim comments on YouTube following the October 7 attack.
  • In Canada, most anti-Muslim and antisemitic hate speech related to the Israel-Hamas conflict draws on ‘classical’ tropes and narratives, rather than focussing on the conflict itself.
  • These include portraying Muslims as violent, or Jews as having disproportionate and insidious power and influence.

Background: Post-7 October Rise in Hate Crime 

The Hamas terrorist attack on Israel on 7 October, followed by the Israel-Hamas war, has led to a significant surge in reported antisemitic and anti-Muslim hate incidents in Canada. Internationally, this trend has also manifested online. Previous ISD research on antisemitic and anti-Muslim speech during the current Israel-Hamas conflict identified a 50-fold increase in the volume of antisemitic comments and a 43-fold increase in the volume of anti-Muslim comments following 7 October compared to the period immediately preceding the Hamas attacks.  

This article builds on previous studies by summarising ISD’s topic modelling analysis of antisemitic and anti-Muslim hate speech on Canadian YouTube. Our focus is a qualitative breakdown of the most important narratives identified by ISD researchers within comments containing antisemitic and anti-Muslim hate speech (see Annex for definitions). 

Approach 

Video Collection 

ISD first collected all videos from 43 popular Canadian channels (see Annex for details on channel selection and filtered out any videos published before October 1 2023. This resulted in 4,818 videos and 827,478 comments between 1 October 2023 and 13 April 2024. These comments were run through existing annotators to create a sub-set of comments that were possibly antisemitic or anti-Muslim (see Annex details on keyword filters).  

This resulted in 29,681 potentially antisemitic and 24,209 potentially anti-Muslim comments (a small number of messages were identified as possibly both antisemitic and anti-Muslim). We subsequently applied a length limit and excluded non-English comments, reducing our dataset to 23,667 potentially antisemitic and 19,255 potentially anti-Muslim comments. Analysts then manually reviewed comments to assess how accurate the keyword filter had been in identifying antisemitic and anti-Muslim comments.  

Topic modelling was then used to automatically identify distinct clusters of messages that occur within a dataset based on their linguistic similarity (‘topics’) within this dataset. Topics were labelled by their relevance based on a random sample of 10-20 related comments and qualitatively grouped into themes (more detail in the methodological Annex).  

From 54 clusters for potentially antisemitic comments, 34 were excluded as either no consistent theme could be identified or the majority of comments were judged to not be antisemitic. The remaining 20 clusters (sub-themes), representing a total of 8,847 comments, were grouped into 7 over-arching master themes (see breakdown in the following section).  

From 50 clusters for potentially anti-Muslim comments, 27 were excluded as either no consistent theme could be identified or the majority of comments were judged not to be anti-Muslim. The remaining 23 clusters (sub-themes), representing a total of 13,344 comments, were grouped into 6 over-arching master themes (see breakdown in the following section).  

It should be noted that while not every comment in the clusters which were grouped into master and sub-themes were hateful, the majority in each cluster were assessed to be antisemitic or anti-Muslim by ISD annotators. Rather than providing an exact assessment of the scale of antisemitic and anti-Muslim comments under Canadian YouTube channels, the aim of this step was to provide a qualitative overview of the relative prevalence of recurring themes within the dataset.  

Qualitative analysis of key topics identified 

The following section outlines the findings of ISD’s qualitative analysis of key topics identified through the topic modelling approach (outlined in detail in the Annex).  

Figure 1: Master themes identified through topic modelling among potentially anti-Muslim comments.

Figure 1: Master themes identified through topic modelling among potentially anti-Muslim comments.

Among the anti-Muslim comments identified on Canadian YouTube channels, allegations that Muslims collectively support terrorism and political violence accounted for nearly three-quarters (72.1 percent) of the comments. Among these were messages which sought to blame Muslims collectively for terrorist groups such as Hamas or describe the actions of such groups as representative of Muslims or the tenets of Islam. Many comments tried to portray Muslim immigrants to Canada as terrorists and raised concerns about their influence over the Trudeau government’s policy towards Israel. Other messages described the doctrines of Islam as evil or inherently violent, for example by mockingly using calling it the “Religion of Peace.”  

The second-most prominent theme among comments classified as anti-Muslim (17.6 percent of the total) were narratives around Muslims being inherently hateful, intolerant or culturally incompatible with Canadian society. For example, such comments would make blanket claims alleging that Muslims or Islam are intrinsically antisemitic. Others described Muslims as hateful of non-Muslims or uncivilised and argued that their values were therefore incompatible with a peaceful coexistence within Canada’s diverse society.  

Smaller numbers of comments were centred around takeover narratives (the idea that Muslim immigration is a demographic threat to Canada), allegations that Muslims are dishonest (e.g. false claims that the doctrine of Taqiyya allows Muslims to lie) or calls for discrimination against or deportations of Canadian Muslims. Some posts also portrayed Muslims as naturally misogynist and opposed to women’s rights.  

Figure 2: Master themes identified through topic modelling among potentially antisemitic comments.

Figure 2: Master themes identified through topic modelling among potentially antisemitic comments.

Among comments on Canadian YouTube channels identified as antisemitic, the biggest group contained conspiracy theories about the malign influence of Jewish power (70.3 percent of the total). This included claims that Jewish NGOs are facilitating mass immigration (including of Muslims) to foment civil unrest in Canada and conspiracy theories about Jewish control of the media or desire to “destroy the West.” Some of these posts were simultaneously antiMuslim and antisemitic. Many posts also argued that Jewish people are and have historically been responsible for major world events such as armed conflicts. Some posts recycled popular antisemitic conspiracy theories around 2001 attack on the World Trade Center or the USS Liberty incident (when Israel mistakenly struck an American warship during the 1967 war). While some comments used the words ‘Zionists’ rather than Jewish people, ISD researchers nevertheless judged that posts alleging that “Zionists control every aspect of our lives” or which talked about “zionist occupied western governments” indisputably reproduce antisemitic stereotypes.  

Comments featuring demonising or dehumanising rhetoric against Jewish people was the second most common type of antisemitism (10.2 percent of the total). Among these comments, ISD researchers identified claims that hatred of Jews was justified due their behaviour or the actions of the government of Israel. Other sub-themes demonised Jewish people or Judaism as satanic or portrayed Jewish people as prone to lying.  

Smaller numbers of comments equated the actions of the Israeli government with that of the Nazis, for example by saying that “israelis can no longer cry against holocaust because they portray themselves now as the jewish zionists/nazis who are causing modern holocaust to palestenian civilians”. Here it is worth highlighting that, in line with previous ISD research on antisemitism online, Israel-related antisemitism makes up a minority of comments in the dataset. This leads ISD to conclude that important definitional debates around drawing thresholds between anti-Israel and antisemitic narratives should not distract from efforts to document and counter the evident rise in antisemitism since 7 October.  

Other comments trivialised or minimised the Holocaust, downplayed antisemitism (“listen to all the whiny jews, oh the victim hood on full display”), or used language associated with classical antisemitism (“Jews killed Jesus”). A small number of posts justified harming Jewish people, including calls for them to be deported from both Canada and Israel.  

Conclusion 

This analysis reveals that in the Canadian context, a majority of both anti-Muslim and antisemitic hate speech relating to the ongoing conflict in Gaza draws on well-trodden tropes. Rather than focusing specifically on dynamics relating to the 7 October attack and the subsequent conflict, the hate speech identified here leans into established themes which seek to paint Muslims as violent, or which draw on conspiracy theories about Jewish power. Importantly, these findings are in line with complementary work produced by ISD looking at global trends in anti-Muslim and antisemitic discourse. These findings are helpful as they suggest that ongoing debates around definitional thresholds may not be particularly relevant in a context where familiar and clear-cut examples of hate speech represent the most hateful discussion. Additionally, these findings may be informative for practitioners seeking to counter the amplification and spread of hate speech online, by demonstrating the most salient trends in discussion.  

Annex: Methodology and Data Collection 

Defining Antisemitism and anti-Muslim Hate 

ISD is informed by  the definition of antisemitism of the International Holocaust Remembrance Alliance (IHRA), a widely accepted and useful definition for measuring the diverse manifestations of contemporary antisemitism. The IHRA definition, and its 11 accompanying examples, are designed as guides for narratives which may, in context, constitute antisemitism. 

Anti-Muslim hate is defined using ISD’s definition of targeted hate as activity which seeks to dehumanise, demonise, harass, threaten, or incite violence against an individual or community based on religion, ethnicity, race, sex, gender identity, sexual orientation, disability, national origin or migrant status”. It was additionally recognised that anti-Muslim hate uses stereotypes or slurs about Muslims or Islam rooted in Islamophobia. 

There have been significant debates around the thresholds for antisemitism and anti-Muslim hate, especially in relation to the line that divides them from legitimate criticism of Israel or Islam. ISD draws on definitions that acknowledge that not every criticism of the actions of Israel is antisemitic and not every criticism of the tenets of Islam is anti-Muslim.  

Using the IHRA definition, we draw a distinction between legitimate criticism of Israel and attacks which hold all Jews accountable for Israel’s actions. It should be noted that several examples listed in the IHRA definition related to Israel have been particularly controversial. The IHRA definition itself notes that “the overall context” should be considered when assessing whether something is antisemitic. In the context of the current Israel-Hamas conflict, accusations of double standards and comparisons with Nazi Germany have been rife. It is therefore critical to differentiate between accusations of war crimes and genocide against Israel that are rooted in international law and a desire to advocate for human rights and self-determination for Palestinians, and comments which diminish or trivialise the Holocaust, or imply a double standard against Jewish people .  

Similarly, there is a clear dividing line between legitimate criticism of Islam and anti-Muslim speech. The later essentialises and homogenises the religion of Islam and, by implication, its adherents. For example, ISD researchers would not code comments criticising specific extremist groups that self-identify as Muslim, or doctrines used to justify violence. On the other hand, describing Islam as a “death cult” or an inherently violent religion would be classified as anti-Muslim.  

Channel selection 

To collate a list of the most popular YouTube channels in Canada, ISD researchers used the online tool Hypeauditor. Based on the data from Hypeauditor, we produced lists of the 50 most subscribed and the 50 most popular YouTube channels in Canada in the ‘News and Politics’ category, ranked by views.  

ISD researchers subsequently reviewed these channels manually and filtered out those which were unlikely to produce content relevant to the Israel-Hamas conflict (e.g. channels focused on entertainment, sports or lifestyle trends). Other YouTube account listing sites, such as Feedspot, were also used to complement ISD’s list. This resulted in an overall list of 43 popular Canadian YouTube channels focused on current events. 

Video Collection 

ISD first collected all videos from the 43 Canadian channels identified and filtered out any videos published before October 1 2023. This resulted in 4,818 videos.  

In total, ISD downloaded 827,478 comments from these 4,818 videos using the YouTube API posted between 1 October 2023 and 13 April 2024. These comments were run through existing annotators to create a sub-set of comments that were possibly antisemitic or anti-Muslim. This filtering process relied on three lists of keywords which were relevant to the current conflict in Israel and Palestine, antisemitism, and anti-Muslim hate. The relevancy filtering combined a mix of exact phrases usually included in hateful posts, such as slurs or known slogans (e.g. ‘without lies islam dies’); and the combination of target group identifier and hateful, violent or stereotypical term (e.g: ‘jew’ + ‘control the media’).  

This resulted in 29,681 potentially antisemitic and 24,209 potentially anti-Muslim comments (a small number of messages were identified as possibly both antisemitic and anti-Muslim).  

Within the overall dataset there was a subset of comments whose length exceeded the limit of what our semantic mapping model could represent. These longer comments were excluded for two reasons: firstly, if left in, the semantic mapping model would truncate the comment after the limit was reached creating data not representative of the full comment. We also found that when comparing the semantic mapping model with and without the truncated comments removed, the former removed produced a much clearer representation of the semantic themes within the dataset. Excluding these comments produced 23,667 potentially antisemitic and 19,255 potentially anti-Muslim comments. 

All comments were additionally put through an English language annotator to identify English messages. All non-English messages were excluded.  

Analysts then manually reviewed comments to assess how accurate the keyword filter had been in identifying antisemitic and anti-Muslim comments. From the sample of 200 potentially antisemitic comments, 25 were labelled as hateful (12.5 percent), while from the sample of 200 potentially anti-Muslim comments, 90 were labelled as hateful (45 percent). Based on this manual review of both potentially antisemitic and anti-Muslim comments, ISD and CASM assessed that our data contained enough hateful messages to use semantic mapping models with the aim of identifying the most prominent hateful narratives within these comments.  

Topic modelling  

Topic modelling is a powerful method for identifying recurring discussions, sentiments and themes in large-scale text data, and patterns in how these discourses are used. Topic modelling was used to analyse the dataset of 23,667 potentially antisemitic and 19,255 potentially anti-Muslim comments from selected Canadian YouTube channels using BERTopic, a machine learning and natural language processing tool. This enabled the automated mapping and identification of ‘topics’, distinct clusters of messages that occur within a dataset based on their linguistic similarity. Topics were labelled by their relevance based on a random sample of 10-20 related comments. Topics were then qualitatively grouped into themes.  

List of Popular Canadian YouTube Channels focussed on Current Events 

CBC Fifth 

The CCF  

JordanBPeterson 

Global News 

CBC News 

CBC The National 

Rebel News Online 

CTV News 

Canada Proud 

City News 

Pierre Poilievre 

The Globe and Mail 

Clyde Do Something 

Michelle Rempel Garner 

Breakfast Television 

CHCH News 

CP24 

Progressive News Network 

The Based Conservative  

TVA Nouvelles 

Canada Stands 

House of Canada 

Mister Sunshine Baby 

CBC Vancouver 

Official W 5  

Toronto Sun 

Street Politics Canada 

Short Fat Otaku 

Northern Perspective 

Canadainfo 

National Post  

Lauren Southern 

cpac 

Toronto Star  

The Pleb reporter 

The Rational National  

Haider Mehdi 

Moose on the Loose 

True North 

TVO Today 

Canadian Press 

CBC NL 

APTN News 

False and unverified claims proliferate online following Trump assassination attempt

Unverified claims about the attempted assassination of former President Donald Trump proliferated across social media in the 24 hours following the incident and continue to spread in the absence of more detailed factual information about the shooter. ISD identified the main false claims being promoted and how they mutated and spread across platforms.